Title: Modeling and Analysis of Wrinkles on Aging Human Faces
The analysis and modeling of aging human faces has been done extensively in the past years regarding several applications. Most of this work is based on maching learning techniques focused on appearance of faces at different ages incorporating facial features such as face shape/geometry and patch-based texture features. However, we do not find much work done on analysis of facial wrinkles explicitly in general and for a person in specific. The goal of this dissertation is to analyse and model facial wrinkles specifically for different applications.
Facial wrinkles are challenging low-level image features to analyse. In general, skin texture has drastically varying appearance due to its characteristic physical properties. A skin patch looks very different when viewed or illuminated from different angles. This makes subtle skin features like facial wrinkles difficult to be detected in images acquired in uncontrolled imaging settings. In this dissertation, we examine the image properties of wrinkles i.e. intensity gradients and geometric
properties and use them for several applications from computer vision to computer graphics. The proposed applications include low-level image processing for automatic detection/localization of wrinkles, soft biometrics and removal of wrinkles using digital inpainting.
First, we present results of detection/localization of wrinkles in images using Marked Point Process (MPP).Wrinkles are modeled as sequences of line segments in a Bayesian framework which incorporates a prior probability model based on the likely geometric properties of wrinkles and a data likelihood term based on image intensity gradients. Wrinkles are localized by sampling the posterior probability using a Reversible Jump Markov Chain Monte Carlo (RJMCMC) algorithm.
We also present an evaluation algorithm to quantitatively evaluate the detection and false alarm rate of our algorithm and conduct experiments with images taken in uncontrolled settings.
The MPP modeling, despite its promising localization results, requires a large number of iterations in the RJMCMC algorithm to reach global minimum resulting in considerable computation time. This motivated us to adopt a deterministic approach based on image morphology for fast localization of facial wrinkles. We propose image features based on Gabor filter bank to highlight subtle curvilinear discontinuities in skin texture caused by wrinkles. Then, image morphology is
used to incorporate geometric constraints to localize curvilinear shapes of wrinkles at image sites of large Gabor filter responses. We conduct experiments on two sets of low and high resolution images to show faster and visually better localization results as compared to those obtained by MPP modeling.
As a next application, we investigate the user-drawn and automatically detected wrinkles as a pattern for their discriminative power as a soft biometrics to recognize subjects from their wrinkle patterns only. A set of facial wrinkles from an image is treated as a curve pattern and used for subject recognition. Given the wrinkle patterns from a query and a gallery image, several distance measures are calculated between the two patterns to quantify the similarity between them. This is done by finding the possible correspondences between curves from the two patterns using a simple bipartite graph matching algorithm. Then several metrics are used to calculate the similarity between the two wrinkle patterns. These metrics are based on Hausdorff distance and curve-to-curve correspondences. We conduct experiments on data sets of both hand drawn and automatically detected wrinkles.
Finally, we apply digital inpainting to automatically remove wrinkles from facial images. Digital image inpainting refers to filling in the holes of arbitrary shapes in images so that they seem to be a part of the original image. The inpainting methods target one or both of the structure and texture of an image. There are two limitations of current inpainting methods for the removal of wrinkles. First, the difference in the attributes of structure and texture requires different inpainting
methods. Facial wrinkles do not fall strictly under the category of structure or texture and can be considered as some where in between. Second, almost all of the image inpainting techniques are supervised i.e. the area/gap to be filled is provided by user interaction and the algorithms attempt to find the suitable image portion automatically. We present an unsupervised image inpainting method where facial regions with wrinkles are detected autmatically using their characteristic
intensity gradients and removed by painting the regions by the surrounding skin texture.